Convert the rhltable to a plain hashtable under a spinlock. The walks
keep their place with a cursor entry in the chain. A claim is unhashed
the moment it's dropped and a walk that finds a superblock that isn't
born yet waits for it to be born or to die and looks again. Since
hashing can't fail anymore the entry for the superblock's own s_dev is
embedded in struct super_block and the preallocation in alloc_super()
goes away. This all simplifies things quite a lot.
It also fixes a btrfs bug where fs_bdev_freeze() ended up freezing a
btrfs filesystem through a stale member that open_ctree() had already
released while the freeze was waiting for SB_BORN. The resume doesn't
find a holder for that device anymore so the filesystem stays frozen
for good.
The walks return superblocks instead of table entries now and
user_get_super() hands out the passive reference the walk took.
Reproducer, a three device btrfs with a stale copy of a removed member.
dm-delay under the remaining members stretches the window between
btrfs_open_devices() and btrfs_free_extra_devids():
truncate -s 1G a1 a2 b
A1=$(losetup -f --show a1); A2=$(losetup -f --show a2); B=$(losetup -f --show b)
mkfs.btrfs -f -d single -m single $A1 $A2 $B
cp b b.old
mount $A1 /mnt && btrfs device remove 3 /mnt && umount /mnt
OLD=$(losetup -f --show b.old)
dmsetup create d1 --table "0 $(blockdev --getsz $A1) delay $A1 0 200"
dmsetup create d2 --table "0 $(blockdev --getsz $A2) delay $A2 0 200"
dmsetup create old --table "0 $(blockdev --getsz $OLD) linear $OLD 0"
mount -o device=/dev/mapper/d2,device=/dev/mapper/old /dev/mapper/d1 /mnt &
sleep 1; dmsetup suspend old; wait
dmsetup resume old
touch /mnt/foo # hangs until fsfreeze -u /mnt
That left btrfs frozen in 5 of 5 rounds with the same tables issued
through raw dm ioctls. The -o degraded variant needs the mount held
between the insertion and the bd_fsfreeze_count check in
fs_bdev_register() with a kprobe.
Fixes: cdb5146f8d ("fs: look up superblocks via the device table in fs_holder_ops")
Link: https://patch.msgid.link/20261012-work-super-dev_hash-v1-2-305cf71cd334@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Please consider pulling these changes from the signed vfs-7.4-rc1.sync_close tag.
Thanks!
Christian
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCatHrlwAKCRCRxhvAZXjc
orrgAQDrCihXTPRcJkiQfDH6LGeRdV+vD0xf2Sp39MMuDOrAggEA+eya4nPNqY8n
K7DK7PJZ/DvyEHckxQQpgmGGqlE4bw0=
=zb7I
-----END PGP SIGNATURE-----
Merge tag 'vfs-7.4-rc1.sync_close' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull file closing updates from Christian Brauner:
"This makes closing files synchronous for close_range(), exec and exit.
Since fput() was switched to task work every final __fput() of an
exiting task is queued on task work and run from exit_task_work().
exec and close_range() do the same. For a task with a large descriptor
table that is a cmpxchg() and a TIF_NOTIFY_RESUME test-and-set per
file plus a list entry for any later task_work_cancel() to search
under pi_lock, and then a second pass over every struct file that
isn't cache hot anymore by then. Oleg and Neil have pointed this out
over the years.
close_files(), close_range() and close-on-exec now flush and put each
file inline the way close(2) does. Kernel threads aren't affected.
They share init_files which init_task pins forever so their
exit_files() never drops the last reference, and their fput() keeps
punting to a workqueue. The one place that could put a table from a
kthread context, a failing copy_process(), can't hold the last
reference while the source table is alive.
On a 64 vCPU KVM guest tearing down 4096, 65536 and 1M descriptors
scattered at random positions in the table:
4096 fds 65536 fds 1M fds
exit 1045 -> 895 us 18.9 -> 14.9 ms 429 -> 306 ms
execve, close-on-exec 1393 -> 1190 us 21.4 -> 16.9 ms 471 -> 325 ms
close_range() 1179 -> 988 us 21.9 -> 17.4 ms 484 -> 337 ms
So that is 16-21% or 40-150 ns per closed file, largest at 1M
descriptors. With eight processes tearing down 65536 descriptors each
at once it stays at 16-21% because the second pass over 16 MB of
struct file is what gets expensive when CPUs compete for the cache.
With 32 or 64 processes both kernels are bound by pushing millions of
files through slab and rcu and the difference shrinks to 4-8%. Files
whose own release dominates gain less, 13% for eventfds, 11% for
pipes, 6% for AF_UNIX sockets. close(2) is untouched and will-it-scale
is within 2%.
The spufs coredump support is removed. It was the only reason
do_coredump() unshared the descriptor table. gdb dropped Cell
debugging in 9.1 and binutils in 2.34, the Cell blades are gone and
nothing on the PlayStation 3 produces or reads these notes. With that
unshare_files() goes away, exec installs its private table through the
new switch_files_struct() and the coredump startup completion becomes
a plain thread count waited on with the new wait_var_event_state().
Three fixes for the synchronous close are included. A failed fork
released its files under scx_fork_rwsem and deadlocked against the bpf
link's release. A session leader's exit released its tty before
disassociate_ctty() ran and the foreground job lost its SIGHUP, so the
tty is hung up before the files are closed now. And descriptor tables
are walked from the highest descriptor down which restores the order
the deferred puts had and keeps a few long-standing release ordering
bugs from deadlocking exit"
* tag 'vfs-7.4-rc1.sync_close' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
fs: close files from the highest descriptor down
coredump: factor out coredump_wait_inactive()
exit: hang up the tty before closing the files
fork: release the files of a failed fork after sched_cancel_fork()
coredump: replace the startup completion with a thread count
sched: add wait_var_event_state()
coredump: drop core_state->dumper
fs: make close_cloexec_files() synchronous
fs: rename do_close_on_exec() to close_cloexec_files()
fs: make close_range() synchronous
fs: make close_files() synchronous
fs: add filp_close_sync()
fs: remove unshare_files()
fs: move unshare_fd() to fs/file.c
fs: add switch_files_struct()
coredump: stop unsharing the file descriptor table
fs: don't open-code file_close_fd() in close_fd()
powerpc: remove coredump support
Please consider pulling these changes from the signed vfs-7.4-rc1.lookup tag.
Thanks!
Christian
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCatHrlgAKCRCRxhvAZXjc
ohALAP95TywGM0X9vNlKadgGMzD4ISQ0lo9xrdWVZ+eKN3DHSwEA/F9DnWdCX+gs
pNVcTf8cqjRUEUSDN8TcLSV12LOIBgc=
=yoIV
-----END PGP SIGNATURE-----
Merge tag 'vfs-7.4-rc1.lookup' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull path lookup updates from Christian Brauner:
"This prepares for Neil's changes to directory locking. The goal is for
the VFS to lock the dentry and not the parent directory and to push
the i_rwsem parent locking down into the filesystems where it can be
kept, removed or adjusted as each filesystem sees fit.
The next step in that is to swap the lock order of d_alloc_parallel()
and i_rwsem so code can wait on i_rwsem while holding an in-lookup
dentry. Two things have to change before that can happen. Nobody may
call d_alloc_parallel() while holding i_rwsem, and nobody may d_drop()
a dentry while it is being worked on since that effectively unlocks an
in-lookup dentry and lets another one be allocated too early.
So d_splice_alias() learns to handle hashed dentries in addition to
unhashed and in-lookup ones, which is what ->atomic_open() needs and
removes the reason filesystems d_drop() first. d_alloc_trylock() is a
non-blocking d_alloc_parallel() for the filesystems that prime the
dcache from readdir, smb, nfs, fuse and ceph, where waiting for a
concurrent lookup is pointless. d_duplicate() creates an in-lookup
dentry with the name of an existing one for silly-rename and
whiteouts. LOOKUP_SHARED tells ->lookup() how the parent is locked so
it can drop and retake it. DCACHE_PAR_LOOKUP is tracked by lockdep now
with a per filesystem type key, and overlayfs gets its own class since
it stacks over itself. And a d_flags bit is reserved for filesystem
private use.
Also included is Mateusz's fix for the spurious dentry reference cycle
on open. __legitimize_path() grabs a reference, do_dentry_open() grabs
another and terminate_walk() drops the first. do_dentry_open()
consumes the held reference now. Opening the same file read-only on a
20-core VM goes from 4,043,375 to 5,629,378 ops/s in will-it-scale"
* tag 'vfs-7.4-rc1.lookup' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
VFS: d_duplicate: repair comments to avoid kernel-doc warnings
ovl: provide private lockdep class for dentry->dep_map
VFS: rename lookup_map to dep_map
VFS: per-fs-type lockdep key for dentry in-lookup locking
VFS: reserve a d_flags bit for fs-specific usage
VFS: add lockdep monitoring of DCACHE_PAR_LOOKUP lock.
VFS: Add LOOKUP_SHARED flag.
VFS: add d_duplicate()
VFS: introduce d_alloc_trylock()
VFS: enhance d_splice_alias() to handle hashed dentries
VFS: fix various typos in documentation for start_creating start_removing etc
fs: avoid spurious dentry ref/unref cycle on open
Currently close() already does a synchronous release of the last
reference since the task is about to return to userspace and the
deferral through task work buys nothing. Add a filp_close_sync() helper.
We'll use that in the next patches.
No functional changes.
Link: https://patch.msgid.link/20260910-work-coredump-unlock-self-v4-5-a5c1800dc930@kernel.org
Reviewed-by: NeilBrown <neil@brown.name>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Now that everything takes a mnt_idmap as const store a const pointer in
struct vfsmount, struct mount_kattr and struct kstatmount and return one
from mnt_idmap(), file_mnt_idmap() and ovl_upper_mnt_idmap(). Finally,
also make mnt_idmap_get() return a const pointer. Also convert the
remaining local variables that are initialized from the accessors.
alloc_mnt_idmap() keeps returning a non-const pointer. It is the only
place where an idmapping is actually written to.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-26-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-25-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-24-54ccd48e100b@kernel.org
Acked-by: Paul Moore <paul@paul-moore.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-23-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-16-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-15-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-12-54ccd48e100b@kernel.org
Acked-by: Paul Moore <paul@paul-moore.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-8-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-5-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Convert to const struct mnt_idmap.
A mount's idmapping is immutable. The only thing that is allowed to be
modified afterwards is the reference count and that is hidden behind
mnt_idmap_get() and mnt_idmap_put(). Everything else only ever reads
from the idmapping. This is the same model that struct cred uses and the
idmapping is also rather sensitive.
So make the idmap argument const wherever we can. The conversion is done
from the bottom up so callers can continue to pass a non-const pointer
to a const parameter until the conversion is finished.
No functional changes.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-3-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
struct mnt_idmap is immutable once alloc_mnt_idmap() has set it up. The
only thing that changes afterwards is the reference count. Model
mnt_idmap_{get,put}() after struct cred's reference count handling and
take a const pointer.
This model is forced upon us because ceph stores the idmapping in mds
requests and need to be able to take and put references.
mnt_idmap_get() keeps returning a non-const pointer for now because
struct vfsmount still stores a non-const pointer. We can only change
that at the end of the series once every user of mnt_idmap() has been
ported.
Link: https://patch.msgid.link/20260901-work-idmap-const-v1-2-54ccd48e100b@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Opening a file grabs a reference on the terminal dentry in
__legitimize_path(), then another one in do_dentry_open() and finally
drops the initial reference in terminate_walk().
That's 2 modifications which don't need to be there -- do_dentry_open()
can consume the already held reference instead.
When benchmarking on a 20-core vm using will-it-scale to open the same
file read-only, the results are (ops/s):
before: 4043375
after: 5629378 (+39%)
Signed-off-by: Mateusz Guzik <mjguzik@gmail.com>
Link: https://patch.msgid.link/20260803125138.1937674-1-mjguzik@gmail.com
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Please consider pulling these changes from the signed vfs-7.3-rc1.super tag.
Thanks!
Christian
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCan7RJQAKCRCRxhvAZXjc
oroBAP40BufiChnC6I9Ev79QWF+Pfri1r/3AR4xXRG4HkXQaYQD/YE+Ria3+Bt35
+tWU66cYlQ/+9LfU8DJ3KXm7jPUBiQA=
=hXLH
-----END PGP SIGNATURE-----
Merge tag 'vfs-7.3-rc1.super' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull vfs superblock updates from Christian Brauner:
- Make it possible to share a block device between multiple
filesystems.
erofs can mount read-only blob devices shared between many
superblocks, but because we only tracked a single superblock a
freeze, thaw, removal or sync on such a device was never propagated
to all the superblocks using it, and there was no way to find them.
Add an efficient table to lookup all superblocks using a given block
device.
- A bunch of pre-existing fixes fell out of this work:
A block-device freeze racing a btrfs device change could leave the
whole filesystem stuck frozen. A bdev_freeze() issued by "dmsetup
suspend" or an LVM snapshot resolves that holder to freeze the
filesystem. and bdev_thaw() resolves it again to thaw. A freeze
landing while btrfs is adding, removing or replacing a device freezes
the filesystem. The membership change then drops that link. So the
matching thaw could no longer find the superblock.
Forbid freezing a device for the duration of a membership change,
modelled on deny_write_access()/allow_write_access().
* tag 'vfs-7.3-rc1.super' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (24 commits)
super: fix dying superblock warning messages
block: reject block device inodes with i_rdev == 0 in lookup_bdev()
selftests/filesystems: add ustat() coverage
fs: look up the superblock via the device table in user_get_super()
super: make fs_holder_ops private
f2fs: open via dedicated fs bdev helpers
erofs: open via dedicated fs bdev helpers
fs: tolerate per-superblock freeze errors on shared devices
fs: look up superblocks via the device table in fs_holder_ops
ext4: open via dedicated fs bdev helpers
btrfs: open via dedicated fs bdev helpers
xfs: port to fs_bdev_file_open_by_path()
fs: add dedicated block device open helpers for filesystems
fs: maintain a global device-to-superblock table
ocfs2: don't reset s_dev on dismount
ext4: use anonymous devices for KUnit test superblocks
fs, block: move blk_mode_t and fop_flags_t into <linux/types.h>
super: take lock after last reference count
super: convert s_count to refcount_t s_passive
btrfs: deny freezing devices undergoing a replace
...
Add a new file descriptor sentinel FD_FAILFS_ROOT following
FD_PIDFS_ROOT and FD_NSFS_ROOT and teach fchdir() to accept it. A
process calling fchdir(FD_FAILFS_ROOT) moves its working directory
into failfs. Every AT_FDCWD-relative lookup afterwards fails with
EOPNOTSUPP including "." and ".." and getcwd() reports the working
directory as unreachable from the process root by returning a path
prefixed with "(unreachable)". Lookups relative to explicit directory
file descriptors are unaffected.
The sentinel is the only way in. No privilege or gating is required.
Setting the working directory to a directory in which every operation
fails grants nothing and loses nothing that closing file descriptors
couldn't lose. An unlinked working directory behaves the same way today
modulo errno. The working directory also plays no role in confining ".."
resolution so no boundary is weakened.
Link: https://patch.msgid.link/20260724-work-failfs-v2-2-485dabbae185@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
nullfs provides a permanently empty and immutable directory. Lookups
fail with ENOENT. The directory can be opened, read, stat, mounted upon.
It behaves like nothing is there.
Add its counterpart failfs where the semantics are not "there is
nothing here" but "nothing is supported here". Every operation that
reaches the filesystem fails with EOPNOTSUPP. Even statfs()/fstatfs()
fail so the filesystem cannot be discovered through an fd to it.
EOPNOTSUPP rather than a permission errno keeps that coherent. There
is no permission model in which anything could ever be allowed and
EACCES or EPERM would merely suggest that different credentials might
succeed while EIO would suggest corruption. It also makes hitting the
failfs boundary mostly quite dinstinguishable. A task anchoring its
lookups at real directory file descriptors may be able to tell a failfs
refusal from an ordinary permission failure. I wouldn't go so far as
guaranteeing that but it should mostly work.
No path lookup can open the root, not even with O_PATH. It is never
reached by a lookup in a parent directory. The only way to a path-walk
terminal at the root is a jump through a /proc/<pid>/{root,cwd} magic
link or by mountpoint traversal. The root also refuses
->d_weak_revalidate() which the VFS calls for jumped terminals. That
covers the jump-based references too: an O_PATH open is refused,
name_to_handle_at() cannot encode it into a file handle, and following a
magic link into it fails. A plain readlink() of such a link still works
and shows "failfs:/".
There is a single instance of failfs mounted during early boot via
kern_mount() making it logically distinct from every mount namespace.
Since the mount is a member of no mount namespace mounting onto it
fails. So nothing can ever be mounted on top of it. It cannot be cloned
via OPEN_TREE_CLONE and it does not show up in statmount()/listmount()
or /proc/<pid>/mountinfo. The filesystem is not registered so it is
not visible in /proc/filesystems and cannot be mounted from userspace.
This lets tasks shed their filesystem state completely. A process with
its root directory or working directory in failfs must anchor every path
lookup at an explicit file descriptor or is doomed to fail any lookup.
Absolute paths, absolute symlinks, and AT_FDCWD-relative lookups
simply fail. Followup patches will expose it via a new FD_FAILFS_ROOT
file descriptor sentinel understood by fchdir() and the new fchroot()
system call.
Link: https://patch.msgid.link/20260724-work-failfs-v2-1-485dabbae185@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
fs_holder_ops recovers the owning superblock from bdev->bd_holder, which
forces the holder to be exactly one superblock and prevents several
superblocks from sharing one block device. That's what erofs is doing.
As a first step introduce a global dev_t-keyed rhltable mapping each
device to the superblock(s) using it. The entry is preallocated in
alloc_super() and registered under sb->s_dev by the set callback through
set_anon_super() and set_bdev_super(), the two helpers every set
callback assigns s_dev through. Registration is the final fallible act
of a set callback, so an insert failure unwinds through sget_fc()'s
existing set-failure path: the fs_context keeps ownership of s_fs_info
and the callers' error paths stay correct. set_anon_super() releases
the anonymous dev it allocated when registration fails. Unwinding
through deactivate_locked_super() instead would run kill_sb() and free
s_fs_info behind the caller's back: nfs and ceph free that object
through a local pointer when sget_fc() fails and would double-free.
The superblock stashes the entry in sb->s_super_dev and
kill_super_notify() drops the claim through it, so teardown doesn't
depend on s_dev staying stable; an entry that was never registered is
freed together with the superblock in destroy_super_work().
Each table entry holds a passive reference (s_passive) on its
superblock, so the struct stays valid for as long as the entry is
reachable. Entries are claim-counted through sd_ref: additional claims
on the same (device, superblock) pair share the entry, and the unlink
is deferred to the last put, so a later iteration cursor never resumes
from a removed node.
The table is initialized from mnt_init(): the first superblocks (the
tmpfs shm mount and rootfs) are created from start_kernel() long before
any initcall runs, so an initcall would be too late.
The table has no readers yet; the fs_holder_ops callbacks are switched
over once all devices a filesystem claims are registered.
Link: https://patch.msgid.link/20260616-work-super-bdev_holder_global-v2-7-7df6b864028e@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
This fixe does page cache invalidation after DIO and async DIO writes for
both O_DIRECT and FOPEN_DIRECT_IO cases.
Commit b359af8275 ("fuse: Invalidate the page cache after FOPEN_DIRECT_IO
write") fixed xfstests generic/209 for DIO writes in the FOPEN_DIRECT_IO
path. DIO writes without FOPEN_DIRECT_IO are already handled by
generic_file_direct_write().
However, async DIO writes (xfstests generic/451) remain unhandled.
After this fix:
- Async write with FUSE_ASYNC_DIO:
invalidate in fuse_aio_invalidate_worker()
- Otherwise (Sync or async write without FUSE_ASYNC_DIO):
- With FOPEN_DIRECT_IO:
invalidate in fuse_direct_write_iter()
- Without FOPEN_DIRECT_IO:
invalidate in generic_file_direct_write()
Workqueue is required for async write invalidation to prevent deadlock:
calling it directly in the I/O end routine (which is in fuse worker thread
context) can block on a folio lock held by a buffered I/O thread waiting
for the same fuse worker thread.
Co-developed-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Signed-off-by: Cheng Ding <cding@ddn.com>
Reviewed-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
-----BEGIN PGP SIGNATURE-----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=giHS
-----END PGP SIGNATURE-----
Merge tag 'lsm-pr-20260410' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/lsm
Pull LSM updates from Paul Moore:
"We only have five patches in the LSM tree, but three of the five are
for an important bugfix relating to overlayfs and the mmap() and
mprotect() access controls for LSMs. Highlights below:
- Fix problems with the mmap() and mprotect() LSM hooks on overlayfs
As we are dealing with problems both in mmap() and mprotect() there
are essentially two components to this fix, spread across three
patches with all marked for stable.
The simplest portion of the fix is the creation of a new LSM hook,
security_mmap_backing_file(), that is used to enforce LSM mmap()
access controls on backing files in the stacked/overlayfs case. The
existing security_mmap_file() does not have visibility past the
user file. You can see from the associated SELinux hook callback
the code is fairly straightforward.
The mprotect() fix is a bit more complicated as there is no way in
the mprotect() code path to inspect both the user and backing
files, and bolting on a second file reference to vm_area_struct
wasn't really an option.
The solution taken here adds a LSM security blob and associated
hooks to the backing_file struct that LSMs can use to capture and
store relevant information from the user file. While the necessary
SELinux information is relatively small, a single u32, I expect
other LSMs to require more than that, and a dedicated backing_file
LSM blob provides a storage mechanism without negatively impacting
other filesystems.
I want to note that other LSMs beyond SELinux have been involved in
the discussion of the fixes presented here and they are working on
their own related changes using these new hooks, but due to other
issues those patches will be coming at a later date.
- Use kstrdup_const()/kfree_const() for securityfs symlink targets
- Resolve a handful of kernel-doc warnings in cred.h"
* tag 'lsm-pr-20260410' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/lsm:
selinux: fix overlayfs mmap() and mprotect() access checks
lsm: add backing_file LSM hooks
fs: prepare for adding LSM blob to backing_file
securityfs: use kstrdup_const() to manage symlink targets
cred: fix kernel-doc warnings in cred.h
Stacked filesystems such as overlayfs do not currently provide the
necessary mechanisms for LSMs to properly enforce access controls on the
mmap() and mprotect() operations. In order to resolve this gap, a LSM
security blob is being added to the backing_file struct and the following
new LSM hooks are being created:
security_backing_file_alloc()
security_backing_file_free()
security_mmap_backing_file()
The first two hooks are to manage the lifecycle of the LSM security blob
in the backing_file struct, while the third provides a new mmap() access
control point for the underlying backing file. It is also expected that
LSMs will likely want to update their security_file_mprotect() callback
to address issues with their mprotect() controls, but that does not
require a change to the security_file_mprotect() LSM hook.
There are a three other small changes to support these new LSM hooks:
* Pass the user file associated with a backing file down to
alloc_empty_backing_file() so it can be included in the
security_backing_file_alloc() hook.
* Add getter and setter functions for the backing_file struct LSM blob
as the backing_file struct remains private to fs/file_table.c.
* Constify the file struct field in the LSM common_audit_data struct to
better support LSMs that need to pass a const file struct pointer into
the common LSM audit code.
Thanks to Arnd Bergmann for identifying the missing EXPORT_SYMBOL_GPL()
and supplying a fixup.
Cc: stable@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-unionfs@vger.kernel.org
Cc: linux-erofs@lists.ozlabs.org
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Serge Hallyn <serge@hallyn.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Paul Moore <paul@paul-moore.com>
Pass the flags one level down to replace the somewhat confusing small
argument, and clean up do_truncate as a result.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260323070205.2939118-3-hch@lst.de
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
The "small" argument to do_sys_ftruncate indicates if > 32-bit size
should be reject, but all the arch-specific compat ftruncate64
implementations get this wrong. Merge do_sys_ftruncate and
ksys_ftruncate, replace the integer as boolean small flag with a
descriptive one about LFS semantics, and use it correctly in the
architecture-specific ftruncate64 implementations.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Fixes: 3dd681d944 ("arm64: 32-bit (compat) applications support")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260323070205.2939118-2-hch@lst.de
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
[mostly] sanitize struct filename hanling
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQqUNBr3gm4hGXdBJlZ7Krx/gZQ6wUCaYlcJgAKCRBZ7Krx/gZQ
6xlKAP9c9J13sJ/mcobsj1Ov7nSHISNbnYqvRRCu09Wq3UQvJgEApNQYOEdLtpff
zUnWOAQ0nOKY7w9VMLkRRustXpuGjAc=
=Fld4
-----END PGP SIGNATURE-----
Merge tag 'pull-filename' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull vfs 'struct filename' updates from Al Viro:
"[Mostly] sanitize struct filename handling"
* tag 'pull-filename' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (68 commits)
sysfs(2): fs_index() argument is _not_ a pathname
alpha: switch osf_mount() to strndup_user()
ksmbd: use CLASS(filename_kernel)
mqueue: switch to CLASS(filename)
user_statfs(): switch to CLASS(filename)
statx: switch to CLASS(filename_maybe_null)
quotactl_block(): switch to CLASS(filename)
chroot(2): switch to CLASS(filename)
move_mount(2): switch to CLASS(filename_maybe_null)
namei.c: switch user pathname imports to CLASS(filename{,_flags})
namei.c: convert getname_kernel() callers to CLASS(filename_kernel)
do_f{chmod,chown,access}at(): use CLASS(filename_uflags)
do_readlinkat(): switch to CLASS(filename_flags)
do_sys_truncate(): switch to CLASS(filename)
do_utimes_path(): switch to CLASS(filename_uflags)
chdir(2): unspaghettify a bit...
do_fchownat(): unspaghettify a bit...
fspick(2): use CLASS(filename_flags)
name_to_handle_at(): use CLASS(filename_uflags)
vfs_open_tree(): use CLASS(filename_uflags)
...
Please consider pulling these changes from the signed vfs-7.0-rc1.namespace tag.
Thanks!
Christian
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCaYX49gAKCRCRxhvAZXjc
ovzgAP9BpqMQhMy2VCurru8/T5VAd6eJdgXzEfXqMksL5BNm8gEAsLx666KJNKgm
Sh/yVA2KBjf51gvcLZ4gHOISaMU8bAI=
=RGLf
-----END PGP SIGNATURE-----
Merge tag 'vfs-7.0-rc1.namespace' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull vfs mount updates from Christian Brauner:
- statmount: accept fd as a parameter
Extend struct mnt_id_req with a file descriptor field and a new
STATMOUNT_BY_FD flag. When set, statmount() returns mount information
for the mount the fd resides on — including detached mounts
(unmounted via umount2(MNT_DETACH)).
For detached mounts the STATMOUNT_MNT_POINT and STATMOUNT_MNT_NS_ID
mask bits are cleared since neither is meaningful. The capability
check is skipped for STATMOUNT_BY_FD since holding an fd already
implies prior access to the mount and equivalent information is
available through fstatfs() and /proc/pid/mountinfo without
privilege. Includes comprehensive selftests covering both attached
and detached mount cases.
- fs: Remove internal old mount API code (1 patch)
Now that every in-tree filesystem has been converted to the new
mount API, remove all the legacy shim code in fs_context.c that
handled unconverted filesystems. This deletes ~280 lines including
legacy_init_fs_context(), the legacy_fs_context struct, and
associated wrappers. The mount(2) syscall path for userspace remains
untouched. Documentation references to the legacy callbacks are
cleaned up.
- mount: add OPEN_TREE_NAMESPACE to open_tree()
Container runtimes currently use CLONE_NEWNS to copy the caller's
entire mount namespace — only to then pivot_root() and recursively
unmount everything they just copied. With large mount tables and
thousands of parallel container launches this creates significant
contention on the namespace semaphore.
OPEN_TREE_NAMESPACE copies only the specified mount tree (like
OPEN_TREE_CLONE) but returns a mount namespace fd instead of a
detached mount fd. The new namespace contains the copied tree mounted
on top of a clone of the real rootfs.
This functions as a combined unshare(CLONE_NEWNS) + pivot_root() in a
single syscall. Works with user namespaces: an unshare(CLONE_NEWUSER)
followed by OPEN_TREE_NAMESPACE creates a mount namespace owned by
the new user namespace. Mount namespace file mounts are excluded from
the copy to prevent cycles. Includes ~1000 lines of selftests"
* tag 'vfs-7.0-rc1.namespace' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
selftests/open_tree: add OPEN_TREE_NAMESPACE tests
mount: add OPEN_TREE_NAMESPACE
fs: Remove internal old mount API code
selftests: statmount: tests for STATMOUNT_BY_FD
statmount: accept fd as a parameter
statmount: permission check should return EPERM
Please consider pulling these changes from the signed vfs-7.0-rc1.nullfs tag.
Thanks!
Christian
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCaYX49gAKCRCRxhvAZXjc
olG7AQD9TywOR0HC9PMT8jrhC1TKODnZ4H1aLNlYVltzfJ09xwEAwFSGO4rQmGAF
aZdD0RQw4bkf7IC1PIZHEGUqmVXJCQ8=
=NvyI
-----END PGP SIGNATURE-----
Merge tag 'vfs-7.0-rc1.nullfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull vfs nullfs update from Christian Brauner:
"Add a completely catatonic minimal pseudo filesystem called "nullfs"
and make pivot_root() work in the initramfs.
Currently pivot_root() does not work on the real rootfs because it
cannot be unmounted. Userspace has to recursively delete initramfs
contents manually before continuing boot, using the fragile
switch_root sequence (overmount + chroot).
Add nullfs, a minimal immutable filesystem that serves as the true
root of the mount hierarchy. The mutable rootfs (tmpfs/ramfs) is
mounted on top of it. This allows userspace to simply:
chdir(new_root);
pivot_root(".", ".");
umount2(".", MNT_DETACH);
without the traditional switch_root workarounds. systemd already
handles this correctly. It tries pivot_root() first and falls back
to MS_MOVE only when that fails.
This also means rootfs mounts in unprivileged namespaces no longer
need MNT_LOCKED, since the immutable nullfs guarantees nothing can be
revealed by unmounting the covering mount.
nullfs is a single-instance filesystem (get_tree_single()) marked
SB_NOUSER | SB_I_NOEXEC | SB_I_NODEV with an immutable empty root
directory. This means sooner or later it can be used to overmount
other directories to hide their contents without any additional
protection needed.
We enable it unconditionally. If we see any real regression we'll
hide it behind a boot option.
nullfs has extensions beyond this in the future. It will serve as a
concept to support the creation of completely empty mount namespaces -
which is work coming up in the next cycle"
* tag 'vfs-7.0-rc1.nullfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
fs: use nullfs unconditionally as the real rootfs
docs: mention nullfs
fs: add immutable rootfs
fs: add init_pivot_root()
fs: ensure that internal tmpfs mount gets mount id zero
When creating containers the setup usually involves using CLONE_NEWNS
via clone3() or unshare(). This copies the caller's complete mount
namespace. The runtime will also assemble a new rootfs and then use
pivot_root() to switch the old mount tree with the new rootfs. Afterward
it will recursively umount the old mount tree thereby getting rid of all
mounts.
On a basic system here where the mount table isn't particularly large
this still copies about 30 mounts. Copying all of these mounts only to
get rid of them later is pretty wasteful.
This is exacerbated if intermediary mount namespaces are used that only
exist for a very short amount of time and are immediately destroyed
again causing a ton of mounts to be copied and destroyed needlessly.
With a large mount table and a system where thousands or ten-thousands
of containers are spawned in parallel this quickly becomes a bottleneck
increasing contention on the semaphore.
Extend open_tree() with a new OPEN_TREE_NAMESPACE flag. Similar to
OPEN_TREE_CLONE only the indicated mount tree is copied. Instead of
returning a file descriptor referring to that mount tree
OPEN_TREE_NAMESPACE will cause open_tree() to return a file descriptor
to a new mount namespace. In that new mount namespace the copied mount
tree has been mounted on top of a copy of the real rootfs.
The caller can setns() into that mount namespace and perform any
additionally required setup such as move_mount() detached mounts in
there.
This allows OPEN_TREE_NAMESPACE to function as a combined
unshare(CLONE_NEWNS) and pivot_root().
A caller may for example choose to create an extremely minimal rootfs:
fd_mntns = open_tree(-EBADF, "/var/lib/containers/wootwoot", OPEN_TREE_NAMESPACE);
This will create a mount namespace where "wootwoot" has become the
rootfs mounted on top of the real rootfs. The caller can now setns()
into this new mount namespace and assemble additional mounts.
This also works with user namespaces:
unshare(CLONE_NEWUSER);
fd_mntns = open_tree(-EBADF, "/var/lib/containers/wootwoot", OPEN_TREE_NAMESPACE);
which creates a new mount namespace owned by the earlier created user
namespace with "wootwoot" as the rootfs mounted on top of the real
rootfs.
Link: https://patch.msgid.link/20251229-work-empty-namespace-v1-1-bfb24c7b061f@kernel.org
Tested-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Aleksa Sarai <cyphar@cyphar.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Suggested-by: Christian Brauner <brauner@kernel.org>
Suggested-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
filename_renameat2() replaces do_renameat2(); unlike the latter,
it does not drop filename references - these days it can be just
as easily arranged in the caller.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
"filp" thing never made sense; seeing that there are exactly 4 callers
in the entire tree (and it's neither exported nor even declared in
linux/*/*.h), there's no point keeping that ugliness.
FWIW, the 'filp' thing did originate in OSD&I; for some reason Tanenbaum
decided to call the object representing an opened file 'struct filp',
the last letter standing for 'position'. In all Unices, Linux included,
the corresponding object had always been 'struct file'...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Instances of struct filename come from names_cachep (via
__getname()). That is done by getname_flags() and getname_kernel()
and these two are the main callers of __getname(). However, there are
other callers that simply want to allocate PATH_MAX bytes for uses that
have nothing to do with struct filename.
We want saner allocation rules for long pathnames, so that struct
filename would *always* come from names_cachep, with the out-of-line
pathname getting kmalloc'ed. For that we need to be able to change the
size of objects allocated by getname_flags()/getname_kernel().
That requires the rest of __getname() users to stop using
names_cachep; we could explicitly switch all of those to kmalloc(),
but that would cause quite a bit of noise. So the plan is to switch
getname_...() to new helpers and turn __getname() into a wrapper for
kmalloc(). Remaining __getname() users could be converted to explicit
kmalloc() at leisure, hopefully along with figuring out what size do
they really want - PATH_MAX is an overkill for some of them, used out
of laziness ("we have a convenient helper that does 4K allocations and
that's large enough, let's use it").
As a side benefit, names_cachep is no longer used outside
of fs/namei.c, so we can move it there and be done with that.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Centralize how we synchronize a lazytime update into the actual on-disk
timestamp into a single helper.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260108141934.2052404-7-hch@lst.de
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Now that the last in-tree filesystem has been converted to the new mount
API, remove all legacy mount API code designed to handle un-converted
filesystems, and remove associated documentation as well.
(The code to handle the legacy mount(2) syscall from userspace is still
in place, of course.)
Tested with an allmodconfig build on x86_64, and a sanity check of an
old mount(2) syscall mount.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Link: https://patch.msgid.link/20251212174403.2882183-1-sandeen@redhat.com
Signed-off-by: Christian Brauner <brauner@kernel.org>
dentries of created objects in dcache (and undo it when removing those).
Reference is grabbed and not released, but it's not actually _stored_
anywhere. That works, but it's hard to follow and verify; among other
things, we have no way to tell _which_ of the increments is intended
to be an unpaired one. Worse, on removal we need to decide whether
the reference had already been dropped, which can be non-trivial if
that removal is on umount and we need to figure out if this dentry is
pinned due to e.g. unlink() not done. Usually that is handled by using
kill_litter_super() as ->kill_sb(), but there are open-coded special
cases of the same (consider e.g. /proc/self).
Things get simpler if we introduce a new dentry flag (DCACHE_PERSISTENT)
marking those "leaked" dentries. Having it set claims responsibility
for +1 in refcount.
The end result this series is aiming for:
* get these unbalanced dget() and dput() replaced with new primitives that
would, in addition to adjusting refcount, set and clear persistency flag.
* instead of having kill_litter_super() mess with removing the remaining
"leaked" references (e.g. for all tmpfs files that hadn't been removed
prior to umount), have the regular shrink_dcache_for_umount() strip
DCACHE_PERSISTENT of all dentries, dropping the corresponding
reference if it had been set. After that kill_litter_super() becomes
an equivalent of kill_anon_super().
Doing that in a single step is not feasible - it would affect too many places
in too many filesystems. It has to be split into a series.
This work has really started early in 2024; quite a few preliminary pieces
have already gone into mainline. This chunk is finally getting to the
meat of that stuff - infrastructure and most of the conversions to it.
Some pieces are still sitting in the local branches, but the bulk of
that stuff is here.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQqUNBr3gm4hGXdBJlZ7Krx/gZQ6wUCaTEq1wAKCRBZ7Krx/gZQ
643uAQC1rRslhw5l7OjxEpIYbGG4M+QaadN4Nf5Sr2SuTRaPJQD/W4oj/u4C2eCw
Dd3q071tqyvm/PXNgN2EEnIaxlFUlwc=
=rKq+
-----END PGP SIGNATURE-----
Merge tag 'pull-persistency' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull persistent dentry infrastructure and conversion from Al Viro:
"Some filesystems use a kinda-sorta controlled dentry refcount leak to
pin dentries of created objects in dcache (and undo it when removing
those). A reference is grabbed and not released, but it's not actually
_stored_ anywhere.
That works, but it's hard to follow and verify; among other things, we
have no way to tell _which_ of the increments is intended to be an
unpaired one. Worse, on removal we need to decide whether the
reference had already been dropped, which can be non-trivial if that
removal is on umount and we need to figure out if this dentry is
pinned due to e.g. unlink() not done. Usually that is handled by using
kill_litter_super() as ->kill_sb(), but there are open-coded special
cases of the same (consider e.g. /proc/self).
Things get simpler if we introduce a new dentry flag
(DCACHE_PERSISTENT) marking those "leaked" dentries. Having it set
claims responsibility for +1 in refcount.
The end result this series is aiming for:
- get these unbalanced dget() and dput() replaced with new primitives
that would, in addition to adjusting refcount, set and clear
persistency flag.
- instead of having kill_litter_super() mess with removing the
remaining "leaked" references (e.g. for all tmpfs files that hadn't
been removed prior to umount), have the regular
shrink_dcache_for_umount() strip DCACHE_PERSISTENT of all dentries,
dropping the corresponding reference if it had been set. After that
kill_litter_super() becomes an equivalent of kill_anon_super().
Doing that in a single step is not feasible - it would affect too many
places in too many filesystems. It has to be split into a series.
This work has really started early in 2024; quite a few preliminary
pieces have already gone into mainline. This chunk is finally getting
to the meat of that stuff - infrastructure and most of the conversions
to it.
Some pieces are still sitting in the local branches, but the bulk of
that stuff is here"
* tag 'pull-persistency' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (54 commits)
d_make_discardable(): warn if given a non-persistent dentry
kill securityfs_recursive_remove()
convert securityfs
get rid of kill_litter_super()
convert rust_binderfs
convert nfsctl
convert rpc_pipefs
convert hypfs
hypfs: swich hypfs_create_u64() to returning int
hypfs: switch hypfs_create_str() to returning int
hypfs: don't pin dentries twice
convert gadgetfs
gadgetfs: switch to simple_remove_by_name()
convert functionfs
functionfs: switch to simple_remove_by_name()
functionfs: fix the open/removal races
functionfs: need to cancel ->reset_work in ->kill_sb()
functionfs: don't bother with ffs->ref in ffs_data_{opened,closed}()
functionfs: don't abuse ffs_data_closed() on fs shutdown
convert selinuxfs
...
The fact that directory operations (create,remove,rename) are protected
by a lock on the parent is known widely throughout the kernel.
In order to change this - to instead lock the target dentry - it is
best to centralise this knowledge so it can be changed in one place.
This patch introduces start_dirop() which is local to VFS code.
It performs the required locking for create and remove. Rename
will be handled separately.
Various functions with names like start_creating() or start_removing_path(),
some of which already exist, will export this functionality beyond the VFS.
end_dirop() is the partner of start_dirop(). It drops the lock and
releases the reference on the dentry.
It *is* exported so that various end_creating etc functions can be inline.
As vfs_mkdir() drops the dentry on error we cannot use end_dirop() as
that won't unlock when the dentry IS_ERR(). For now we need an explicit
unlock when dentry IS_ERR(). I hope to change vfs_mkdir() to unlock
when it drops a dentry so that explicit unlock can go away.
end_dirop() can always be called on the result of start_dirop(), but not
after vfs_mkdir(). After a vfs_mkdir() we still may need the explicit
unlock as seen in end_creating_path().
As well as adding start_dirop() and end_dirop()
this patch uses them in:
- simple_start_creating (which requires sharing lookup_noperm_common()
with libfs.c)
- start_removing_path / start_removing_user_path_at
- filename_create / end_creating_path()
- do_rmdir(), do_unlinkat()
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: NeilBrown <neil@brown.name>
Link: https://patch.msgid.link/20251113002050.676694-3-neilb@ownmail.net
Tested-by: syzbot@syzkaller.appspotmail.com
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQqUNBr3gm4hGXdBJlZ7Krx/gZQ6wUCaN3daAAKCRBZ7Krx/gZQ
6zNWAP9kD6rOJRNqDgea4pibDPa47Tps/WM5tsDv3dsLliY29gEA6sveOWZ3guAj
4oY3ts/NtHLWXvhI7Vd/1mr2aTKEZQk=
=YNK+
-----END PGP SIGNATURE-----
Merge tag 'pull-f_path' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull file->f_path constification from Al Viro:
"Only one thing was modifying ->f_path of an opened file - acct(2).
Massaging that away and constifying a bunch of struct path * arguments
in functions that might be given &file->f_path ends up with the
situation where we can turn ->f_path into an anon union of const
struct path f_path and struct path __f_path, the latter modified only
in a few places in fs/{file_table,open,namei}.c, all for struct file
instances that are yet to be opened"
* tag 'pull-f_path' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (23 commits)
Have cc(1) catch attempts to modify ->f_path
kernel/acct.c: saner struct file treatment
configfs:get_target() - release path as soon as we grab configfs_item reference
apparmor/af_unix: constify struct path * arguments
ovl_is_real_file: constify realpath argument
ovl_sync_file(): constify path argument
ovl_lower_dir(): constify path argument
ovl_get_verity_digest(): constify path argument
ovl_validate_verity(): constify {meta,data}path arguments
ovl_ensure_verity_loaded(): constify datapath argument
ksmbd_vfs_set_init_posix_acl(): constify path argument
ksmbd_vfs_inherit_posix_acl(): constify path argument
ksmbd_vfs_kern_path_unlock(): constify path argument
ksmbd_vfs_path_lookup_locked(): root_share_path can be const struct path *
check_export(): constify path argument
export_operations->open(): constify path argument
rqst_exp_get_by_name(): constify path argument
nfs: constify path argument of __vfs_getattr()
bpf...d_path(): constify path argument
done_path_create(): constify path argument
...
* saner handling of guards in fs/namespace.c, getting
rid of needlessly strong locking in some of the users.
* lock_mount() calling conventions change - have it set
the environment for attaching to given location, storing the
results in caller-supplied object, without altering the passed
struct path. Make unlock_mount() called as __cleanup for those
objects. It's not exactly guard(), but similar to it.
* MNT_WRITE_HOLD done right - mnt_hold_writers() does *not*
mess with ->mnt_flags anymore, so insertion of a new mount into
->s_mounts of underlying superblock does not, in itself, expose
->mnt_flags of that mount to concurrent modifications.
* getting rid of pathological cases when umount() spends
quadratic time removing the victims from propagation graph -
part of that had been dealt with last cycle, this should finish
it.
* a bunch of stuff constified.
* assorted cleanups.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQqUNBr3gm4hGXdBJlZ7Krx/gZQ6wUCaNhzLAAKCRBZ7Krx/gZQ
63/IAP4yxJ6e3Pt66Uw0MeuSNmeLsQwb7mYo72lsYHpxjYANZAEAspMaLDU9NHxM
Dy6WDVoJnf7+aDlD6E443YMfPX8XRQM=
=5T+t
-----END PGP SIGNATURE-----
Merge tag 'pull-mount' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull vfs mount updates from Al Viro:
"Several piles this cycle, this mount-related one being the largest and
trickiest:
- saner handling of guards in fs/namespace.c, getting rid of
needlessly strong locking in some of the users
- lock_mount() calling conventions change - have it set the
environment for attaching to given location, storing the results in
caller-supplied object, without altering the passed struct path.
Make unlock_mount() called as __cleanup for those objects. It's not
exactly guard(), but similar to it
- MNT_WRITE_HOLD done right.
mnt_hold_writers() does *not* mess with ->mnt_flags anymore, so
insertion of a new mount into ->s_mounts of underlying superblock
does not, in itself, expose ->mnt_flags of that mount to concurrent
modifications
- getting rid of pathological cases when umount() spends quadratic
time removing the victims from propagation graph - part of that had
been dealt with last cycle, this should finish it
- a bunch of stuff constified
- assorted cleanups
* tag 'pull-mount' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (64 commits)
constify {__,}mnt_is_readonly()
WRITE_HOLD machinery: no need for to bump mount_lock seqcount
struct mount: relocate MNT_WRITE_HOLD bit
preparations to taking MNT_WRITE_HOLD out of ->mnt_flags
setup_mnt(): primitive for connecting a mount to filesystem
simplify the callers of mnt_unhold_writers()
copy_mnt_ns(): use guards
copy_mnt_ns(): use the regular mechanism for freeing empty mnt_ns on failure
open_detached_copy(): separate creation of namespace into helper
open_detached_copy(): don't bother with mount_lock_hash()
path_has_submounts(): use guard(mount_locked_reader)
fs/namespace.c: sanitize descriptions for {__,}lookup_mnt()
ecryptfs: get rid of pointless mount references in ecryptfs dentries
umount_tree(): take all victims out of propagation graph at once
do_mount(): use __free(path_put)
do_move_mount_old(): use __free(path_put)
constify can_move_mount_beneath() arguments
path_umount(): constify struct path argument
may_copy_tree(), __do_loopback(): constify struct path argument
path_mount(): constify struct path argument
...
Pidfd file handles are exhaustive meaning they don't require a handle on
another pidfd to pass to open_by_handle_at() so it can derive the
filesystem to decode in. Instead it can be derived from the file
handle itself. The same is possible for namespace file handles.
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
